SC02: Bayesian Study Design and Analysis in Regulatory Science

The Reverend Thomas Bayes, a scholar born 300 years ago, and his paper first published in 1763 gradually drew widespread attention in the statistics field. Proponents of Bayesian methods argue that, compared to traditional "frequentist" statistical methods, Bayesian approaches offer significant advantages. They point out that expressing scientific results within the Bayesian framework not only enhances their interpretability but also reduces the likelihood of misinterpreting borderline or inconclusive findings. In Bayesian analysis, a set of historical/external observations is regarded as evidence that updates prior beliefs. Bayesian advocates further highlight that these methods can accelerate the process of drug/medical product clinical trials, make them more equitable, and enhance computational efficiency. However, in the field of regulatory science for drug/medical product, although discussions around Bayesian methods are common, their practical applications remain relatively limited. To this end, we aim to provide a comprehensive course to participants with the essential skills for Bayesian design and data analysis in clinical trials, especially within the context of regulatory science. This course will not only cover the core theoretical foundations of Bayesian methods but also delve into practical implementation, case studies, and communication. Through this short course, participants will be better prepared to tackle complex regulatory challenges and fully leverage the potential of Bayesian methods in clinical trials. This short course features the following three parts: 1) The first part (instructor from academia) Bayesian sample size determination (SSD) has a long history. The early work on Bayesian SSD can be traced to the 1990s. Recently, several new methods on Bayesian designs of clinical trials have been developed with a focus on controlling type I error and power. In the first part of this short course, an overview of the literature on Bayesian SSD will be provided. The general theory and various methods of Bayesian SSD will be presented. Informative prior elicitations to leverage external or historical data will be discussed in detail. The recent development of software to implement various Bayesian SSD approaches will also be reviewed. The first part starts with a brief review of early development of Bayesian SSD. Then, a comprehensive review of Bayesian methods for borrowing historical information and proper use of these methods in Bayesian clinical trial designs follows. This part of the short course will also cover the computational algorithms and recent available software on Bayesian SSD. The first part will also highlight several important applications in designing clinical trials to demonstrate the superiority of Bayesian SSD. 2) The second part (instructor from industry) Following trial design and sample size calculation and completion of trial conduct, the next critical step in clinical trials is the analysis and reporting phase. Bayesian methodologies have gained traction in regulatory submissions due to their flexibility in integrating diverse data sources while addressing two key scientific questions: interpolation and extrapolation. This course segment will focus on essential aspects of Bayesian analysis and reporting in these contexts. Topics include evaluating data sources for fit-for-purpose use, determining the ordering and weighting of evidence, and linking analytical results to target product profile endpoints. We will explore Bayesian model selection, control of Type I error, effective sample size estimation, and sensitivity analyses. Special emphasis will be placed on managing multiple endpoints within interpolation and extrapolation frameworks, as well as conducting subgroup analyses and standardization procedures. The course will conclude by highlighting best practices for generating clear, interpretable labeling language based on Bayesian inference, ensuring scientific rigor and regulatory compliance. 3) The third part (instructors from regulatory agencies including CDER, CDRH, and CBER) The use of Bayesian statistics to support regulatory evaluation of medical devices began in the late 1990s. This segment of the short course gives an overview from a regulatory perspective of Bayesian adaptive design, a study design strategy that uses Bayesian techniques to guide clinical trial decisions based on accumulating data. Predictive probability can be used to make enrollment decisions. Posterior distributions quantify the uncertainty about the parameters of interest, and thus may serve as the basis for declaring study success. At the design stage, it is critical to control type 1 error rate and ensure adequate power, typically via simulation. Regulatory considerations like these will be discussed in realistic settings. In the CDER, Bayesian methods are seeing increasing usage in several areas, including notably pediatric extrapolation, where Bayesian methods are used to borrow information from trials conducted in other populations (such as adults or older pediatrics). In situations like this where borrowing is used, it is necessary to construct an informative prior that represents the consensus on the degree of support for the question of interest in the target population. General considerations for the use of informative priors will be discussed from a regulatory perspective.